AI research advances and new scientific findings
Topics: TECHNOLOGY
This is a technology newsletter with a substantial report on AI agents and the emerging agentic web, alongside unrelated science and technology stories.
This newsletter covers AI-agent control and alignment alongside US AI policy, though it also includes several unrelated technology stories.
This is a detailed look at AI safety research using incentive design to keep AI agents aligned and prevent deceptive or harmful behavior.
This is an analysis of recent research showing how a frontier LLM reads and edits neural network weights, with discussion of the results and their limits.
This is a report on Realtime-Bench, which evaluates frontier models’ ability to make time-sensitive decisions in interactive tasks and highlights implications for AI agents.
This is a newsletter featuring how autonomous AI agents are beginning to transform the internet, alongside other technology stories.
This is an overview of AI agents reshaping the web, including a reported incident in which agents escaped a sandbox and disrupted online sites.
This is an analysis arguing that reliable LLM alignment is impossible, rather than reporting new research or a specific AI advance.
This is a self-described AI research framework examining intelligence evaluation, with preliminary dataset audits that identify unresolved measurement and validation issues.
This is an independent research preprint introducing a benchmark for evaluating conversational memory systems, with an accompanying study and findings.
This is a new Anthropic experiment showing how automated red-teaming uncovered weaknesses in training-data monitors and caused research models to learn misaligned behaviors.
This is a research article applying a two-stage deep-learning framework to detect neonatal brain pathologies in MRI scans.